Optimising Laser Powder Bed Fusion to identify stable manufacturing regimes

Advisors: Dr. Sameh Tawfick and Dr. Elif Ertekin

January 2022 - present

• Built unsupervised image analysis techniques to help segmentation of cross-section images of single-track melt pools.
• Trained a neural network to perform the segmentation automatically with an accuracy greater than 99% and extract the melt pool features.
• Using Bayesian optimisation to exploit the experimental results and find the region of the parameter space in conduction mode, while employing the normalised enthalpy to transition across different materials systems.
• Published software (Python package) for image analysis of melt pool cross-sections.
• Working in collaboration with MIT and NIST to develop standards for the industry.

Links to the software:

Graphene Recipes for Synthesis of High-Quality Materials (Gr-ResQ)

Advisors: Dr. Sameh Tawfick and Dr. Elif Ertekin

September 2019 - present

• Built a chemical vapour deposition system and synthesised graphene. Varied specific reaction parameters - such as total gas flow rate and growth duration - using design of experiments and achieved high repeatability in the quality of graphene.
• Performed active learning using Bayesian optimisation to exploit the experimental results and facilitate more efficient discovery of complex synthesis recipes.
• Designed a custom kernel for a Gaussian process model to simulate the CVD process and explain the balance between the governing physical processes.
• Published software on nanoHUB to enable crowd-sourcing of synthesis recipes and analysis of microscopy images and Raman spectra, with more than 1000 total users worldwide.
• Automated segmentation of scanning electron microscopy images of graphene with a deep neural network with 94.5% pixel-wise accuracy using only 93 training images. Published the trained model, compatible with deepImageJ, at doi.org/10.5281/zenodo.7063245.

Links to the software:

Statistical modelling for design optimisation & performance validation of the Fuse series printers

Manager: Ryan Buck, Optics and Thermal Systems team, Formlabs Inc.

September 2023 - December 2023

• Developed a Standard Operating Procedure (SOP) for collecting seed data using design of experiments to support a data-driven model.
• Built an active-learning model to predict the EAB, UTS, and modulus, along with experimental uncertainty as a function of optical and thermal inputs for the SLS printers for PA12. Developed tailored acquisition functions to guide the print setting optimisation based on special requirements.
• This model explained the variability across printers, and is guiding optimisation efforts for the next generation of printers.
• The method, agnostic of the material system provided to it, is also being used to build similar models for other materials with little-to-no tuning by the engineers.
• Characterised thermal non-uniformity within the print volume and built a predictive model for the same as a function of the optical and thermal inputs.